Time Series, Index Numbers and Forecasting

Introduction

Business decisions are rarely based only on current observations. Human resource managers examine employee turnover over several years, compare wage levels across locations, study seasonal absenteeism, and forecast future labour requirements. These tasks require methods that account for change over time and differences in prices, quantities, or values.

A time series is a set of observations recorded in chronological order. Index numbers express relative change in a variable or group of variables, usually compared with a base period. Forecasting uses historical and other relevant information to estimate future outcomes. Together, these methods support workforce planning, budgeting, compensation analysis, demand estimation, and performance management.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain the nature and components of time series.
2. Calculate and interpret moving averages and trend equations.
3. Distinguish between price, quantity, value, and special-purpose index numbers.
4. Construct simple and weighted index numbers.
5. Explain the uses, assumptions, and limitations of forecasting methods.
6. Apply time-series and index-number techniques to HR and business decisions.
7. Evaluate forecast accuracy and recognise ethical risks in quantitative decision-making.

1. Time Series: Meaning and Structure

A time series may be represented as Y_t, where t denotes time and Y_t is the observed value at time t. Examples include monthly recruitment, annual labour costs, weekly sales, quarterly absenteeism, and daily stock prices.

Time-series data differ from cross-sectional data. Cross-sectional data compare different units at one point in time, whereas time-series data observe one phenomenon over several periods. Chronological order is essential because observations may be dependent: employee turnover this month may be related to turnover in previous months.

A useful time series is often analysed through four components:

Two common models are:

Additive model

Multiplicative model

The additive model is appropriate when seasonal fluctuations are approximately constant in absolute size. The multiplicative model is more suitable when seasonal effects grow or shrink in proportion to the level of the series. In the multiplicative model, components are usually expressed as ratios or percentages, with a neutral index of 1 or 100%.

2. Measuring Trend

Trend analysis identifies the underlying long-term movement after short-term fluctuations are reduced. Three elementary approaches are graphical inspection, moving averages, and least-squares trend fitting.

2.1 Graphical Method

Plotting observations against time can reveal upward or downward movement, seasonal repetition, abrupt changes, and outliers. Although visual analysis is useful as a first step, it is subjective and does not provide a precise forecasting equation.

2.2 Moving Averages

A moving average replaces each observation with the average of neighbouring observations. For an odd number k of periods, the simple moving average centred at time t is:

where k = 2r + 1, Y_t is the observation at time t, and MA_t is the smoothed value. Every observation receives equal weight. The first and last r observations cannot receive centred averages unless another convention is used.

For an even number of periods, such as four quarters, the moving averages must normally be centred by averaging two consecutive moving averages. This is necessary because the first four-quarter average lies between periods rather than exactly on one period.

Worked example. Monthly recruitment over five months was 18, 22, 20, 25, and 27 employees. The three-month moving average for month 3 is:

The moving average smooths short-term variation but is not automatically a forecast. It is a descriptive estimate of the local level, and it loses observations at the ends of the series.

2.3 Least-Squares Trend

A linear trend estimates the relationship between the variable and time:

where Ŷ_t is the estimated value, a is the intercept, b is the estimated change per time unit, and t is coded time. The slope is:

and the intercept is:

Here n is the number of observations. These equations minimise the sum of squared deviations between actual and fitted values. If time is coded so that ∑t = 0, then a = ∑Y_t/n and b = ∑tY_t/∑t².

Worked example. Suppose annual training hours for three years are 40, 44, and 46. Code time as −1, 0, and 1. Then a = (40+44+46)/3 = 43.33. Also, ∑tY_t = (−1)(40)+(0)(44)+(1)(46)=6 and ∑t²=2. Therefore b=6/2=3. The trend equation is:

For the next year, t=2, so the forecast is 43.33 + 3(2) = 49.33 training hours. This forecast assumes that the linear trend continues and that no major structural change occurs.

2.4 Seasonal Indices

A seasonal index measures the typical seasonal level relative to the trend or overall average. Under a multiplicative interpretation, an index of 1.20 means that the season is typically 20% above the underlying level, while 0.85 means 15% below it. For quarterly data, seasonal indices commonly sum to 4; for monthly data, they commonly sum to 12, subject to the method used.

A deseasonalised observation is:

The units remain those of the original variable. Seasonal adjustment should not be used to conceal genuine changes in workforce conditions.

3. Forecasting

Forecasting is the systematic estimation of a future value or event. A forecast is not a certainty; it is conditional on assumptions, information, and model choice. Forecasting methods include qualitative judgement, time-series methods, causal models, and scenario analysis.

3.1 Forecasting Approaches

Qualitative methods use expert judgement, interviews, Delphi panels, and managerial consensus. They are valuable when historical data are limited or when a new policy or technology makes the past a poor guide.

Time-series methods use historical patterns, including moving averages, exponential smoothing, and trend models.

Causal methods explain an outcome using predictor variables. For example, labour demand may depend on sales, production volume, and wage rates. A simple regression model is:

where X is the explanatory variable and b estimates the expected change in Y for a one-unit change in X. Causal interpretation requires sound theory and careful control of confounding variables; correlation alone does not prove causation.

3.2 Exponential Smoothing

Simple exponential smoothing updates a forecast using the most recent error:

where F_t+1 is the next forecast, Y_t is the latest actual value, F_t is the current forecast, and α is the smoothing constant with 0 < α < 1. A high α responds rapidly to recent changes; a low α produces greater smoothing. Simple exponential smoothing is appropriate for a series with a relatively stable level and no strong trend or seasonality. Trend-adjusted or seasonal methods are needed when those features are present.

Worked example. If the current forecast is 100 employees, the actual value is 112, and α=0.30, then:

3.3 Forecast Accuracy

Forecasts should be evaluated using data not used to estimate the model, when possible. Common measures include:

MAD is measured in the original units, MSE gives greater weight to large errors, and MAPE is expressed as a percentage. MAPE is unsuitable when actual values are zero and can be unstable when actual values are close to zero. Forecast accuracy should be compared across equivalent periods and business conditions.

4. Index Numbers

An index number expresses relative change rather than absolute magnitude. If a wage rises from 20,000 to 22,000, the current-period index with the earlier wage as base is 110, meaning a 10% increase.

The basic individual index is:

where p₀ is the base-period value and p₁ is the current-period value. The index equals 100 in the base period; values above 100 indicate increases and values below 100 indicate decreases.

4.1 Types of Index Numbers

The value index is:

where p represents price or rate and q represents quantity; subscripts 0 and 1 denote base and current periods.

4.2 Simple Aggregate Price Index

A simple aggregate price index is:

It is easy to calculate but ignores the relative importance of products. Adding prices of items with different units can also be conceptually weak.

4.3 Weighted Index Numbers

A weighted index gives greater importance to items with greater economic significance.

Laspeyres price index:

It uses base-period quantities as weights. It is relatively easy to construct but may overstate the cost increase because it does not reflect substitution toward cheaper goods.

Paasche price index:

It uses current-period quantities and may understate the increase if consumers shift toward cheaper alternatives.

Fisher’s ideal index:

It is the geometric mean of the Laspeyres and Paasche indices. Its use is justified when both price and quantity information are available and a balanced measure is desired.

4.4 Worked Index Example

Assume the following annual compensation basket:

Then:

Therefore:

The weighted wage index is 110, indicating a 10% increase in the cost of the base-period compensation structure.

5. Quality of Index Numbers

A sound index requires a clearly defined objective, appropriate base period, representative items, reliable prices and quantities, consistent units, and transparent weighting. The base period should be reasonably normal; a crisis year may distort comparisons. Base periods may be fixed or updated through chain indices. In a chain index, consecutive-period changes are linked rather than comparing every period with one distant base.

Important limitations include changes in product quality, new goods, substitution, changing employee composition, sampling error, and revisions to source data. An index is an average summary and may not represent every employee or household. A wage index may rise while real purchasing power falls if consumer prices rise faster.

The approximate real wage relationship is:

If the nominal wage index is 110 and the price index is 105, the real wage index is approximately (110/105)×100 = 104.76, suggesting an increase of about 4.76% relative to the base period.

6. Applied Case: Workforce Planning in a Retail Organisation

This is an illustrative case.

Background

A retail organisation has recorded quarterly recruitment for three years. Recruitment is usually higher in the final quarter because of holiday demand. The HR director must estimate next year’s recruitment requirement and assess whether wage growth has kept pace with consumer prices.

Problem

A simple annual average would ignore the recurring fourth-quarter increase. In addition, comparing total payroll alone would confuse wage-rate changes with changes in the number of employees.

Analysis and Approach

The analyst first plots quarterly recruitment and identifies a seasonal pattern. A four-quarter moving average is used to smooth seasonal variation, and seasonal indices are estimated from the historical data. The analyst then fits a trend to deseasonalised recruitment and multiplies the projected trend by the appropriate quarterly seasonal index.

For compensation, the analyst constructs a Laspeyres wage index using the previous year’s employee mix as weights. The result is compared with the consumer price index. If the wage index is 108 and the price index is 112, the approximate real wage index is:

Thus, nominal wages increased, but purchasing power fell by approximately 3.57% relative to the base period.

Outcome

The organisation prepares different staffing scenarios: a baseline forecast, a high-demand forecast, and a low-demand forecast. HR budgets for recruitment costs using the high-demand scenario while retaining flexibility through temporary staffing. Compensation discussions focus on real wages rather than nominal wages alone.

Learning Points

1. Seasonality should not be mistaken for long-term growth.
2. Forecasts should support scenarios rather than create false certainty.
3. Weighted indices are more informative than unweighted totals when categories differ in importance.
4. Nominal increases do not necessarily imply improved employee welfare.
5. HR decisions require both statistical evidence and contextual judgement.

7. Professional, Ethical and Technological Considerations

Forecasts influence hiring, redundancy, promotion, and compensation. Poor data or inappropriate models can therefore create material harm. HR professionals should document data sources, assumptions, transformations, model selection, and uncertainty. Protected characteristics must not be used improperly, and historical discrimination should not be reproduced merely because it appears in past data.

Automated forecasting and artificial-intelligence systems can process large workforce datasets, detect seasonal patterns, and generate scenarios. However, they may amplify bias, obscure assumptions, or create privacy risks. Good governance requires data minimisation, access controls, audit trails, human review, explainability appropriate to the decision, and monitoring for changes in model performance. Forecasts should inform—not replace—professional and employee-sensitive judgement.

8. Best-Practice Workflow

A defensible analysis normally follows these steps:

1. Define the decision, variable, time horizon, and unit of measurement.
2. Collect reliable, comparable, chronologically ordered data.
3. Inspect graphs and summary statistics for trend, seasonality, outliers, and missing values.
4. Select a method consistent with the data structure and purpose.
5. Estimate the model using transparent calculations.
6. Validate accuracy using holdout data or rolling-origin evaluation.
7. Compare alternative models and document assumptions.
8. Present a point forecast together with ranges or scenarios.
9. Monitor actual outcomes and revise the model when conditions change.
10. Review ethical, privacy, and fairness implications before implementation.

Challenges and Limitations

Time-series methods assume that useful patterns in the past will persist. Structural breaks, economic shocks, new technology, changes in labour law, mergers, and pandemics may invalidate that assumption. Short series provide weak evidence, while very long series may combine incompatible regimes. Missing data and inconsistent definitions can produce misleading trends. Index numbers depend on basket design and weights; different legitimate choices may yield different results. Finally, numerical precision does not eliminate uncertainty. A forecast of 1,248 employees should not be interpreted as more certain than a forecast of approximately 1,250.

Chapter Summary

Time series organise observations chronologically and can be analysed through trend, seasonal, cyclical, and irregular components. Moving averages smooth fluctuations, while least-squares equations estimate a trend. Exponential smoothing updates forecasts by combining recent actual values with previous forecasts. Forecast accuracy can be assessed using error, MAD, MSE, and MAPE, although each measure has limitations.

Index numbers measure relative change. Price, quantity, value, wage, and cost-of-living indices serve different purposes. Weighted indices such as Laspeyres, Paasche, and Fisher’s index are more informative than simple aggregates when items differ in importance. Nominal values should be distinguished from real values by considering price changes.

In HRM, these methods support workforce planning, budgeting, compensation analysis, and scenario development. Sound practice requires appropriate data, explicit assumptions, validation, ethical governance, and continuing review. Forecasts are decision aids, not guarantees.

Key Terms

Time series: Chronologically ordered observations of a variable.

Trend: Long-term direction in a time series.

Seasonal variation: Regular movement recurring within a year.

Cyclical variation: Multi-year movement associated with broader economic conditions.

Irregular variation: Unpredictable movement caused by unusual events.

Additive model: Model in which components are combined by addition.

Multiplicative model: Model in which components are combined by multiplication.

Moving average: Average calculated over successive overlapping periods.

Least squares: Estimation method that minimises squared deviations.

Exponential smoothing: Forecasting method that weights recent observations more heavily.

Forecast: Systematic estimate of a future value or event.

Forecast error: Difference between an actual value and its forecast.

Index number: Relative measure comparing a value with a reference period.

Base period: Period assigned an index value of 100.

Price index: Measure of relative price change.

Quantity index: Measure of relative quantity change.

Laspeyres index: Weighted index using base-period quantities.

Paasche index: Weighted index using current-period quantities.

Fisher index: Geometric mean of Laspeyres and Paasche indices.

Real wage: Wage adjusted for changes in prices.

Review Questions

Short Answer Questions

1. What is a time series, and how does it differ from cross-sectional data?
2. Distinguish between seasonal and cyclical variation.
3. Why are moving averages used?
4. What does an index number of 115 indicate?
5. State one limitation of MAPE.

Long Answer Questions

1. Explain the four components of a time series and compare additive and multiplicative models.
2. Derive and explain the meaning of the linear least-squares trend equation.
3. Explain Laspeyres, Paasche, and Fisher price indices, including their advantages and limitations.
4. Discuss the role of forecasting in human resource planning.
5. Explain how ethical and privacy concerns affect automated workforce forecasting.

Analytical / Application Questions

1. A company records monthly absenteeism and observes a regular increase every December. Recommend an analytical procedure for forecasting next year’s monthly absenteeism and justify each step.
2. Nominal wages increase by 8%, while the consumer price index increases by 11%. Calculate the approximate real wage index using a base value of 100 and interpret the result.
3. An HR analyst reports a highly accurate forecast based on historical data, but the organisation has just introduced a four-day workweek. Evaluate whether the forecast should be used without modification and propose a better approach.

Multiple Choice Questions with Answers

Question 1

Which component represents a regular pattern that repeats within a year?

A. Trend

B. Seasonal variation

C. Cyclical variation

D. Irregular variation

Correct Answer: B. Seasonal variation

Question 2

Three-month recruitment observations are 22, 20, and 25. What is their simple moving average?

A. 20.00

B. 21.00

C. 22.33

D. 25.00

Correct Answer: C. 22.33

Question 3

In the trend equation Ŷ_t = a + bt, what does b represent?

A. The estimated value at time zero

B. The number of observations

C. The average seasonal index

D. The estimated change per time unit

Correct Answer: D. The estimated change per time unit

Question 4

If a price rises from 50 to 55, what is its price index using the original price as the base?

A. 90

B. 100

C. 105

D. 110

Correct Answer: D. 110

Question 5

Which index uses base-period quantities as weights?

A. Laspeyres price index

B. Paasche price index

C. Fisher price index only

D. Simple aggregate quantity index

Correct Answer: A. Laspeyres price index

Question 6

If α=0.30, the latest actual value is 112, and the previous forecast is 100, what is the next exponential-smoothing forecast?

A. 100.3

B. 103.6

C. 106.0

D. 112.0

Correct Answer: B. 103.6

Question 7

Which forecast accuracy measure is expressed in the original units of the variable?

A. MAD

B. MAPE

C. Percentage bias

D. Seasonal index

Correct Answer: A. MAD

Question 8

A real wage index is approximately calculated as:

A. Price index ÷ nominal wage index × 100

B. Nominal wage index + price index

C. Nominal wage index ÷ price index × 100

D. Nominal wage index × price index

Correct Answer: C. Nominal wage index ÷ price index × 100

Question 9

Simple exponential smoothing is most appropriate when a series has:

A. No observations

B. A stable level without strong trend or seasonality

C. Only cyclical variation with no level

D. A guaranteed structural break

Correct Answer: B. A stable level without strong trend or seasonality

Question 10

Why should a forecast be accompanied by assumptions or scenarios?

A. Because forecasts are conditional estimates rather than certainties

B. Because historical data are never useful

C. Because indices cannot be calculated

D. Because moving averages always overstate demand

Correct Answer: A. Because forecasts are conditional estimates rather than certainties